-
Notifications
You must be signed in to change notification settings - Fork 0
/
ctc_conv_corrected.py
314 lines (258 loc) · 10.3 KB
/
ctc_conv_corrected.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
# Compatibility imports
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0'
import time
import tensorflow as tf
import scipy.io.wavfile as wav
import numpy as np
from six.moves import xrange as range
try:
from python_speech_features import mfcc, logfbank
except ImportError:
print("Failed to import python_speech_features.\n Try pip install python_speech_features.")
raise ImportError
from utils import sparse_tuple_from as sparse_tuple_from
from utils import pad_sequences as pad_sequences
#import tensorflow.contrib.eager as tfe
#tfe.enable_eager_execution()
SPACE_TOKEN = '<space>'
SPACE_INDEX = 0
FIRST_INDEX = ord('a') - 1 # 0 is reserved to space
# Some configs
num_features = 26
# Accounting the 0th indice + space + blank label = 28 characters
num_classes = ord('z') - ord('a') + 1 + 1 + 1
# Hyper-parameters
num_epochs = 1000
#num_epochs = 100
num_hidden = 256
num_norm_layers = 4
num_resi_layers = 10
batch_size = 32
initial_learning_rate = 1e-5
momentum = 0.9
#XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
# Loading the data
monsterDict = dict()
audioValFN = list()
#ll = os.listdir('data/processed_files/')
with open('vox/gg.txt','r') as fp:
cc = fp.readlines()
chekL = [c.strip('\n') for c in cc]
with open('vox/keyFile.txt','r') as fp:
cc1 = fp.readlines()
textDict = dict()
neoaudioFN = list()
for elem in cc1:
loc = elem.strip('\n').split('#')
if loc[0] in chekL:
textDict[ loc[0] ] = loc[1]
neoaudioFN.append( loc[0] )
print('OOOOOOOOOOOLLLLLLLLLLLLLLLLLLL '+str(len(neoaudioFN)))
print('OOOOOOOOOOOLLLLLLLLLLLLLLLLLLL '+str(len(textDict)))
num_batches_per_epoch = max(int(len(neoaudioFN)/batch_size), 1 )
num_examples = len(neoaudioFN)
print( num_batches_per_epoch )
print( num_examples )
original = ''
batch_train_ip = list()
batch_train_tgt = list()
batch_val_ip = list()
batch_val_tgt = list()
batch_train_seq_len = list()
def retPro( ln ):
z1 = ln.replace('0','zero ')
z2 = z1.replace('1','one ')
z3 = z2.replace('2','two ')
z4 = z3.replace('3','three ')
z5 = z4.replace('4','four ')
z6 = z5.replace('5','five ')
z7 = z6.replace('6','six ')
z8 = z7.replace('7','seven ')
z9 = z8.replace('8','eight ')
z10 = z9.replace('9','nine ')
return z10
def get_a_cell(lstm_size, keep_prob, mode):
lstm = tf.nn.rnn_cell.BasicLSTMCell(lstm_size)
drop = tf.nn.rnn_cell.DropoutWrapper(lstm, output_keep_prob=keep_prob)
if mode == 'RESI':
lstm = tf.contrib.rnn.ResidualWrapper( lstm )
return lstm
return drop
def retFiles( fname, ctr , mode='NA'):
if mode == 'VAL':
fs, audio = wav.read( fname+'.wav' , ctr )
else:
fs, audio = wav.read( 'vox/'+fname+'.wav' , ctr )
inputs = logfbank(audio, samplerate=fs)
#inputs = mfcc(audio, samplerate=fs)
# Tranform in 3D array
train_inputs = inputs
#train_inputs = np.asarray(inputs[np.newaxis, :])
train_inputs = (train_inputs - np.mean(train_inputs))/np.std(train_inputs)
train_seq_len = [train_inputs.shape[1]]
targets = []
# Readings targets
if ctr >= 0:
if mode == 'VAL':
with open( fname+'.txt', 'r') as f:
line = f.readlines()[-1]
proL = retPro( line )
else:
proL = textDict[ fname ]
original = ' '.join(proL.strip().lower().split(' ')[2:]).replace('.', '')
targets = original.replace(' ', ' ')
targets = targets.split(' ')
# Adding blank label
targets = np.hstack([SPACE_TOKEN if x == '' else list(x) for x in targets])
hapur = targets
# Transform char into index
targets = np.asarray([SPACE_INDEX if x == SPACE_TOKEN else ord(x) - FIRST_INDEX
for x in targets])
for elem in targets:
if elem < 0:
print('emir')
print(targets)
print(hapur)
# Creating sparse representation to feed the placeholder
#train_targets = sparse_tuple_from([targets])
train_targets = (targets)
else:
train_targets = []
monsterDict[ fname ] = {'inp': train_inputs ,'tgt': train_targets, 'len': train_seq_len}
return train_inputs, train_targets, train_seq_len
for ctr in range(len(neoaudioFN)):
train_inputs, train_targets, train_seq_len = retFiles( neoaudioFN[ctr] , ctr )
batch_train_ip.append( train_inputs )
batch_train_tgt.append( monsterDict[ neoaudioFN[ctr] ]['tgt'] )
#XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
# You can preprocess the input data here
train_inputs = np.asarray( batch_train_ip )
# You can preprocess the target data here
train_targets = np.asarray( batch_train_tgt )
# THE MAIN CODE!
# Create some wrappers for simplicity
def conv2d(x, W, b, strides=1):
# Conv2D wrapper, with bias and relu activation
print('HOLA '+str(x))
x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='SAME')
x = tf.nn.bias_add(x, b)
return tf.nn.relu(x)
def maxpool2d(x, k=2):
# MaxPool2D wrapper
return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],
padding='SAME')
# Store layers weight & bias
weights = {
# 5x5 conv, 1 input, 32 outputs
'wc1': tf.Variable(tf.random_normal([10, 10, 1, num_classes])),
# 5x5 conv, 32 inputs, 64 outputs
'wc2': tf.Variable(tf.random_normal([8, 8, num_classes, 64])),
# fully connected, 7*7*64 inputs, 1024 outputs
'wd1': tf.Variable(tf.random_normal([num_features*num_classes, 1024])),
# 1024 inputs, 10 outputs (class prediction)
'out': tf.Variable(tf.random_normal([1024, num_classes]))
}
biases = {
'bc1': tf.Variable(tf.random_normal([num_classes])),
'bc2': tf.Variable(tf.random_normal([64])),
'bd1': tf.Variable(tf.random_normal([1024])),
'out': tf.Variable(tf.random_normal([num_classes]))
}
# tf Graph input
X = tf.placeholder(tf.float32, [ None, None, num_features])
Y = tf.sparse_placeholder(tf.int32 )
seq_len = tf.placeholder(tf.int32)
keep_prob = tf.placeholder(tf.float32) # dropout (keep probability)
batch_s = tf.shape(X)[0]
# Construct model
x = tf.reshape(X, shape=[-1, num_features, 399 , 1])
# Convolution Layer
conv1 = conv2d(conv1, weights['wc1'], biases['bc1'], 1)
print('xxxxxxxxxxxxxxxx')
fc1 = tf.reshape(conv1, [batch_size , 399 , weights['wd1'].get_shape().as_list()[0]])
fc1 = tf.layers.dense( fc1, 1024 , activation=tf.nn.relu)
#print('xxxxxxxxxxxxxxxx')
#print(fc1)
#fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
#print('xxxxxxxxxxxxxxxx')
#print(fc1)
# Apply Dropout
fc1 = tf.nn.dropout(fc1, keep_prob)
# Output, class prediction
logits = tf.layers.dense(inputs=fc1, units=num_classes, activation=tf.nn.relu)
logits = tf.transpose(logits, (1, 0, 2))
loss = tf.nn.ctc_loss(Y , logits, seq_len)
cost = tf.reduce_mean(loss)
optimizer = tf.train.MomentumOptimizer(initial_learning_rate,
0.9).minimize(cost)
# Option 2: tf.nn.ctc_beam_search_decoder
# (it's slower but you'll get better results)
decoded, log_prob = tf.nn.ctc_greedy_decoder(logits, seq_len)
# Inaccuracy: label error rate
ler = tf.reduce_mean(tf.edit_distance(tf.cast(decoded[0], tf.int32),
Y))
with tf.Session() as session:
saver = tf.train.Saver()
# Initializate the weights and biases
tf.global_variables_initializer().run()
#saver.restore( session, '/home/ec2-user/speech/tf/resi_save/baby_steps-91' )
for curr_epoch in range(num_epochs):
train_cost = train_ler = 0
start = time.time()
if curr_epoch!=0 and curr_epoch%15 == 0:
initial_learning_rate = initial_learning_rate * 10
momentum = max(0.99, momentum*1.05)
for batch in range(num_batches_per_epoch):
# Getting the index
indexes = [i % num_examples for i in range(batch * batch_size, (batch + 1) * batch_size)]
batch_train_inputs = train_inputs[indexes]
# Padding input to max_time_step of this batch
batch_train_inputs, batch_train_seq_len = pad_sequences(batch_train_inputs)
# Converting to sparse representation so as to to feed SparseTensor input
batch_train_targets = sparse_tuple_from(train_targets[indexes])
feed = {X: batch_train_inputs,
Y: batch_train_targets,
seq_len: batch_train_seq_len,
keep_prob:0.5}
ll, ll1, ll2 = session.run([conv1, fc1, logits], feed)
print( np.asarray(ll).shape )
print( np.asarray(ll1).shape )
print( np.asarray(ll2).shape )
batch_cost, _ = session.run([cost, optimizer], feed)
train_cost += batch_cost*batch_size
train_ler += session.run(ler, feed_dict=feed)*batch_size
# Shuffle the data
shuffled_indexes = np.random.permutation(num_examples)
train_inputs = train_inputs[shuffled_indexes]
train_targets = train_targets[shuffled_indexes]
# Metrics mean
train_cost /= num_examples
train_ler /= num_examples
saved_path = saver.save( session, 'resi_save/baby_steps', global_step=curr_epoch)
log = "Epoch {}/{}, train_cost = {:.3f}, train_ler = {:.3f}, time = {:.3f}"
print(log.format(curr_epoch+1, num_epochs, train_cost, train_ler, time.time() - start))
# Decoding all at once. Note that this isn't the best way
# Padding input to max_time_step of this batch
indexes = [i % num_examples for i in range(batch * batch_size, (batch + 1) * batch_size)]
batch_train_inputs = train_inputs[indexes]
# Padding input to max_time_step of this batch
batch_train_inputs, batch_train_seq_len = pad_sequences(batch_train_inputs)
# Converting to sparse representation so as to to feed SparseTensor input
batch_train_targets = sparse_tuple_from(train_targets[indexes])
feed = {X: batch_train_inputs,
Y: batch_train_targets,
seq_len: batch_train_seq_len,
keep_prob:0.5
}
# Decoding
d = session.run(decoded[0], feed_dict=feed)
dense_decoded = tf.sparse_tensor_to_dense(d, default_value=-1).eval(session=session)
for i, seq in enumerate(dense_decoded):
seq = [s for s in seq if s != -1]
print('Sequence %d' % i)
print('\t Original:\n%s' % train_targets[i])
print('\t Decoded:\n%s' % seq)